Implement function calls on ARM using the blx instruction when available. Usi...

Implement function calls on ARM using the blx instruction when available. Using blx will allow the CPU to predict the return address fo the function, resulting in better overall performamce.

BUG=648

[Erik Corry, 2010-03-19 10:01:05]

LGTM!  I'll land this for you if you fix the minor points below.

http://codereview.chromium.org/1113002/diff/1/3
File src/arm/assembler-arm-inl.h (right):

http://codereview.chromium.org/1113002/diff/1/3#newcode240
src/arm/assembler-arm-inl.h:240: // what need to be patched.
need -> needs

http://codereview.chromium.org/1113002/diff/1/4
File src/arm/assembler-arm.cc (right):

http://codereview.chromium.org/1113002/diff/1/4#newcode244
src/arm/assembler-arm.cc:244: const Instr kLdrPCPattern = al | B26 | L |
pc.code() * B16;
I can't understand why these two aren't divided up into a mask and a value.  Are
there no should-be-zero bits in an LDR pc instruction or a BLX instruction and
if so isn't that just a coincidence?

[rodolph, 2010-03-19 11:39:18]

http://codereview.chromium.org/1113002/diff/1/3
File src/arm/assembler-arm-inl.h (right):

http://codereview.chromium.org/1113002/diff/1/3#newcode240
src/arm/assembler-arm-inl.h:240: // what need to be patched.
On 2010/03/19 10:01:06, Erik Corry wrote:
> need -> needs

Done.

http://codereview.chromium.org/1113002/diff/1/4
File src/arm/assembler-arm.cc (right):

http://codereview.chromium.org/1113002/diff/1/4#newcode244
src/arm/assembler-arm.cc:244: const Instr kLdrPCPattern = al | B26 | L |
pc.code() * B16;
On 2010/03/19 10:01:06, Erik Corry wrote:
> I can't understand why these two aren't divided up into a mask and a value. 
Are
> there no should-be-zero bits in an LDR pc instruction or a BLX instruction and
> if so isn't that just a coincidence?

You are right, these should be divided into mask/value. Done in the latest
upload.

[Erik Corry, 2010-03-19 14:05:48]

Committed to bleeding edge as r4198

On 2010/03/19 11:39:18, rodolph wrote:
> http://codereview.chromium.org/1113002/diff/1/3
> File src/arm/assembler-arm-inl.h (right):
> 
> http://codereview.chromium.org/1113002/diff/1/3#newcode240
> src/arm/assembler-arm-inl.h:240: // what need to be patched.
> On 2010/03/19 10:01:06, Erik Corry wrote:
> > need -> needs
> 
> Done.
> 
> http://codereview.chromium.org/1113002/diff/1/4
> File src/arm/assembler-arm.cc (right):
> 
> http://codereview.chromium.org/1113002/diff/1/4#newcode244
> src/arm/assembler-arm.cc:244: const Instr kLdrPCPattern = al | B26 | L |
> pc.code() * B16;
> On 2010/03/19 10:01:06, Erik Corry wrote:
> > I can't understand why these two aren't divided up into a mask and a value. 
> Are
> > there no should-be-zero bits in an LDR pc instruction or a BLX instruction
and
> > if so isn't that just a coincidence?
> 
> You are right, these should be divided into mask/value. Done in the latest
> upload.