ARM: Emit RSB (reverse-subtract) for subtraction with constant lhs

Emit RSB for subtraction with constant lhs on ARM

Instead of generating the following for "100 - x":

  mov r1, #100
  subs r0, r1, r0

this patch makes Crankshaft generate this:

  rsbs r0, r0, #100

thus saving a register and a mov.

BUG=

Committed: https://code.google.com/p/v8/source/detail?r=13017

[hans, 2012-11-15 16:14:58]

I couldn't measure any performance impact on the benchmarks with this, but
thought I might as well upload it anyway.

Running Octane in the simulator with this patch hits 300k new RST instructions,
so it does occur in benchmarks but perhaps not enough to make a measurable
difference.

I'll leave it to the reviewers to decide whether the improved generated code
quality is worth the changes to V8.

[hans, 2012-11-16 13:58:40]

> this patch makes Crankshaft generate this:
> 
> rsbs r0, r1, #100

Sorry, that should be "rsbs r0, r0, #100"

> thus saving a mov.

and a register.

[ulan_google, 2012-11-20 08:53:06]

Looks good, one small comment:

https://codereview.chromium.org/11308040/diff/1/src/arm/lithium-arm.cc
File src/arm/lithium-arm.cc (right):

https://codereview.chromium.org/11308040/diff/1/src/arm/lithium-arm.cc#newcod...
src/arm/lithium-arm.cc:1356: ASSERT(instr->left().IsInteger32());
This should be ASSERT(instr->left()->representation().IsInteger32()).

[hans, 2012-11-20 10:10:19]

Uploaded new patch.

This one removes the assert that was failing. I've failed
http://code.google.com/p/v8/issues/detail?id=2420 for that.

https://codereview.chromium.org/11308040/diff/1/src/arm/lithium-arm.cc
File src/arm/lithium-arm.cc (right):

https://codereview.chromium.org/11308040/diff/1/src/arm/lithium-arm.cc#newcod...
src/arm/lithium-arm.cc:1356: ASSERT(instr->left().IsInteger32());
On 2012/11/20 08:53:06, ulan1 wrote:
> This should be ASSERT(instr->left()->representation().IsInteger32()).

Done.

[ulan, 2012-11-20 14:45:30]

LGTM, landing