Fix the full compiler on ARM to always generate the same code

Fix the full compiler on ARM to always generate the same code
regardless of the detected CPU.  This is a requirement for the
debugger and the deoptimizer, which both expect that code from
the snapshot (compiled without VFP and ARM7) should have the
same layout as code compiled later.

This is another change to make snapshots more robust with
arbitrary code.
Committed: https://code.google.com/p/v8/source/detail?r=12287

[Yang, 2012-08-10 09:57:11]

LGTM once comments are addressed.

Is a similar change necessary for Intel platforms as well?

https://chromiumcodereview.appspot.com/10824235/diff/1/src/arm/code-stubs-arm.cc
File src/arm/code-stubs-arm.cc (right):

https://chromiumcodereview.appspot.com/10824235/diff/1/src/arm/code-stubs-arm...
src/arm/code-stubs-arm.cc:1945: HeapNumber::kExponentOffset >>
HeapNumber::kMantissaBitsInTopWord;
kExponentOffset's unit is bytes (8 on 32-bit system), kMantissaBitsInTopWord
(20).  The result would be 0.

Don't you mean kNaNOrInfinityLowerBoundUpper32 here? I guess it would be easier
to read if we simply mask the exponent bits and compare to
kNaNOrInfinityLowerBoundUpper32.

https://chromiumcodereview.appspot.com/10824235/diff/1/src/arm/macro-assemble...
File src/arm/macro-assembler-arm.h (right):

https://chromiumcodereview.appspot.com/10824235/diff/1/src/arm/macro-assemble...
src/arm/macro-assembler-arm.h:116: Condition cond = al);
"Size" occurs twice in this method name. Intentional?

[Erik Corry, 2012-08-10 12:24:18]

https://chromiumcodereview.appspot.com/10824235/diff/1/src/arm/code-stubs-arm.cc
File src/arm/code-stubs-arm.cc (right):

https://chromiumcodereview.appspot.com/10824235/diff/1/src/arm/code-stubs-arm...
src/arm/code-stubs-arm.cc:1945: HeapNumber::kExponentOffset >>
HeapNumber::kMantissaBitsInTopWord;
On 2012/08/10 09:57:11, Yang wrote:
> kExponentOffset's unit is bytes (8 on 32-bit system), kMantissaBitsInTopWord
> (20).  The result would be 0.
> 
> Don't you mean kNaNOrInfinityLowerBoundUpper32 here? I guess it would be
easier
> to read if we simply mask the exponent bits and compare to
> kNaNOrInfinityLowerBoundUpper32.

Good catch.  Changed to kExponentMask.

https://chromiumcodereview.appspot.com/10824235/diff/1/src/arm/macro-assemble...
File src/arm/macro-assembler-arm.h (right):

https://chromiumcodereview.appspot.com/10824235/diff/1/src/arm/macro-assemble...
src/arm/macro-assembler-arm.h:116: Condition cond = al);
On 2012/08/10 09:57:11, Yang wrote:
> "Size" occurs twice in this method name. Intentional?

Changed from PredictableSize to PredictableCodeSize to make the connection to
predictable_code_size() more clear.