Optimize the assembly code generated for Math.random()

Optimize the assembly code generated for Math.random()

Committed: http://code.google.com/p/v8/source/detail?r=4384

[William Hesse, 2010-04-09 12:41:56]

I still need to add the instructions to the x64 disassembler.

[Rico, 2010-04-12 07:19:32]

LGTM

http://codereview.chromium.org/1631008/diff/1/4
File src/assembler.h (right):

http://codereview.chromium.org/1631008/diff/1/4#newcode401
src/assembler.h:401: static ExternalReference
fill_heap_number_with_random_function();
Do we still need this?

http://codereview.chromium.org/1631008/diff/1/6
File src/x64/assembler-x64.cc (right):

http://codereview.chromium.org/1631008/diff/1/6#newcode2388
src/x64/assembler-x64.cc:2388: void Assembler::movd(XMMRegister dst, const
Register src) {
I don't think we normally use const on src registers?

http://codereview.chromium.org/1631008/diff/1/7
File src/x64/assembler-x64.h (right):

http://codereview.chromium.org/1631008/diff/1/7#newcode1108
src/x64/assembler-x64.h:1108: // The first argument is the reg field, the second
argument is the r/m field.
I do not get this comment - how can an XMMRegister be a memory field?

http://codereview.chromium.org/1631008/diff/1/8
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/1631008/diff/1/8#newcode4105
src/x64/codegen-x64.cc:4105: ASSERT (CpuFeatures::IsSupported(SSE2));
no space after ASSERT

[William Hesse, 2010-04-12 10:07:27]

http://codereview.chromium.org/1631008/diff/1/4
File src/assembler.h (right):

http://codereview.chromium.org/1631008/diff/1/4#newcode401
src/assembler.h:401: static ExternalReference
fill_heap_number_with_random_function();
On 2010/04/12 07:19:32, Rico wrote:
> Do we still need this?
The ARM platform still uses this.  I have created a bug, requesting that the ARM
platform be changed to work the same way as ia32 and x64.

http://codereview.chromium.org/1631008/diff/1/6
File src/x64/assembler-x64.cc (right):

http://codereview.chromium.org/1631008/diff/1/6#newcode2388
src/x64/assembler-x64.cc:2388: void Assembler::movd(XMMRegister dst, const
Register src) {
On 2010/04/12 07:19:32, Rico wrote:
> I don't think we normally use const on src registers?

Done.

http://codereview.chromium.org/1631008/diff/1/7
File src/x64/assembler-x64.h (right):

http://codereview.chromium.org/1631008/diff/1/7#newcode1108
src/x64/assembler-x64.h:1108: // The first argument is the reg field, the second
argument is the r/m field.
On 2010/04/12 07:19:32, Rico wrote:
> I do not get this comment - how can an XMMRegister be a memory field?

The r/m field contains either a register reference (mode 3) or a memory operand
reference (modes 0, 1, and 2).  If it contains a register reference, it refers
to either an XMM register or a GP register depending on the opcode.

http://codereview.chromium.org/1631008/diff/1/8
File src/x64/codegen-x64.cc (right):

http://codereview.chromium.org/1631008/diff/1/8#newcode4105
src/x64/codegen-x64.cc:4105: ASSERT (CpuFeatures::IsSupported(SSE2));
On 2010/04/12 07:19:32, Rico wrote:
> no space after ASSERT

The ASSERT and Scope for SSE2 are not needed - x64 always has SSE2.

[Rico, 2010-04-12 10:17:30]

http://codereview.chromium.org/1631008/diff/1/7
File src/x64/assembler-x64.h (right):

http://codereview.chromium.org/1631008/diff/1/7#newcode1108
src/x64/assembler-x64.h:1108: // The first argument is the reg field, the second
argument is the r/m field.
On 2010/04/12 10:07:27, William Hesse wrote:
> On 2010/04/12 07:19:32, Rico wrote:
> > I do not get this comment - how can an XMMRegister be a memory field?
> 
> The r/m field contains either a register reference (mode 3) or a memory
operand
> reference (modes 0, 1, and 2).  If it contains a register reference, it refers
> to either an XMM register or a GP register depending on the opcode.
> 
Sorry my bad - I did not realize that the comment where describing the next
three lines (I just thought it was describing the next line)