First step in allocating objects in generated code on ARM

First step in allocating objects in generated code on ARM

This change moves the allocation of new objects into generated code. The allocation will bail out into the runtime system if the number of properties to allocate for the object exceeds the number of in-object properties.

Committed: http://code.google.com/p/v8/source/detail?r=2797

[Erik Corry, 2009-09-01 12:29:36]

LGTM

http://codereview.chromium.org/175045/diff/1/3
File src/arm/builtins-arm.cc (right):

http://codereview.chromium.org/175045/diff/1/3#newcode152
Line 152: __ LoadRoot(r6, Heap::kEmptyFixedArrayRootIndex);
If you move this up then the load has time to complete before the register is
used.

http://codereview.chromium.org/175045/diff/1/3#newcode192
Line 192: __ ldrb(r6, FieldMemOperand(r2, Map::kInObjectPropertiesOffset));
kInObjectPropertiesOffset and kPreAllocatedPropertyFieldsOffset are in the same
word.  This is why they never should have added byte load instructions to the
ARM (cue 4-Yorkshiremen routine...)

http://codereview.chromium.org/175045/diff/1/4
File src/arm/macro-assembler-arm.cc (right):

http://codereview.chromium.org/175045/diff/1/4#newcode809
Line 809: void MacroAssembler::AllocateObjectInNewSpace(Register object_size,
Object size should be in words, rather than bytes.  That saves a shift
instruction.

http://codereview.chromium.org/175045/diff/1/2
File src/arm/macro-assembler-arm.h (right):

http://codereview.chromium.org/175045/diff/1/2#newcode211
Line 211: // it will no longer be allocated. Make sure that no pointers are left
to the
Make -> The caller must make

http://codereview.chromium.org/175045/diff/1/2#newcode213
Line 213: // un-done.
un-done -> undone

http://codereview.chromium.org/175045/diff/1/2#newcode241
Line 241: // Compare instance type in a map.  mpa contains a valid map object
whose
mpa -> Map

[Søren Thygesen Gjesse, 2009-09-01 14:26:13]

http://codereview.chromium.org/175045/diff/1/3
File src/arm/builtins-arm.cc (right):

http://codereview.chromium.org/175045/diff/1/3#newcode152
Line 152: __ LoadRoot(r6, Heap::kEmptyFixedArrayRootIndex);
On 2009/09/01 12:29:36, Erik Corry wrote:
> If you move this up then the load has time to complete before the register is
> used.

Done.

http://codereview.chromium.org/175045/diff/1/3#newcode192
Line 192: __ ldrb(r6, FieldMemOperand(r2, Map::kInObjectPropertiesOffset));
On 2009/09/01 12:29:36, Erik Corry wrote:
> kInObjectPropertiesOffset and kPreAllocatedPropertyFieldsOffset are in the
same
> word.  This is why they never should have added byte load instructions to the
> ARM (cue 4-Yorkshiremen routine...)

Changed to load the instance sizes word and mask/shift to get to the two bytes.

http://codereview.chromium.org/175045/diff/1/4
File src/arm/macro-assembler-arm.cc (right):

http://codereview.chromium.org/175045/diff/1/4#newcode809
Line 809: void MacroAssembler::AllocateObjectInNewSpace(Register object_size,
On 2009/09/01 12:29:36, Erik Corry wrote:
> Object size should be in words, rather than bytes.  That saves a shift
> instruction.

Done, changed the object_size parameter for the other overloaded
AllocateObjectInNewSpace to be words as well for consistency.

http://codereview.chromium.org/175045/diff/1/2
File src/arm/macro-assembler-arm.h (right):

http://codereview.chromium.org/175045/diff/1/2#newcode211
Line 211: // it will no longer be allocated. Make sure that no pointers are left
to the
On 2009/09/01 12:29:36, Erik Corry wrote:
> Make -> The caller must make

Done.

http://codereview.chromium.org/175045/diff/1/2#newcode213
Line 213: // un-done.
On 2009/09/01 12:29:36, Erik Corry wrote:
> un-done -> undone

Done.

http://codereview.chromium.org/175045/diff/1/2#newcode241
Line 241: // Compare instance type in a map.  mpa contains a valid map object
whose
On 2009/09/01 12:29:36, Erik Corry wrote:
> mpa -> Map

Done.

[Erik Corry, 2009-09-02 11:55:07]

LGTM