[turbofan] add some simplifications in the machine operator reducer

[turbofan] add some simplifications in the machine operator reducer

Simplify the '+', '-' and '%' operators with some basic constants (0, 1 or NaN).

R=titzer@google.com, bmeurer@chromium.org

BUG=

[titzer, 2014-09-25 09:08:41]

titzer@chromium.org changed reviewers:
+ titzer@chromium.org

[titzer, 2014-09-25 09:08:41]

https://codereview.chromium.org/599383002/diff/1/src/compiler/machine-operato...
File src/compiler/machine-operator-reducer.cc (right):

https://codereview.chromium.org/599383002/diff/1/src/compiler/machine-operato...
src/compiler/machine-operator-reducer.cc:352: return Replace(m.left().node()); 
// x - 0 => x
Is this correct if left is -0?

https://codereview.chromium.org/599383002/diff/1/test/cctest/compiler/test-ma...
File test/cctest/compiler/test-machine-operator-reducer.cc (right):

https://codereview.chromium.org/599383002/diff/1/test/cctest/compiler/test-ma...
test/cctest/compiler/test-machine-operator-reducer.cc:168: if (tmp.IsSpecial()
&& !tmp.IsInfinite()) {
I would prefer to have this check in the callers.

[vincent.belliard, 2014-09-25 09:51:19]

https://codereview.chromium.org/599383002/diff/1/src/compiler/machine-operato...
File src/compiler/machine-operator-reducer.cc (right):

https://codereview.chromium.org/599383002/diff/1/src/compiler/machine-operato...
src/compiler/machine-operator-reducer.cc:352: return Replace(m.left().node()); 
// x - 0 => x
> Is this correct if left is -0?
Yes. the result is unchanged (except that NaN are normalized on Arm and Arm64).
On Arm and Arm64, we use a subtraction by zero to normalize NaNs.
asm.js use a subtraction by zero to force a number.

https://codereview.chromium.org/599383002/diff/1/test/cctest/compiler/test-ma...
File test/cctest/compiler/test-machine-operator-reducer.cc (right):

https://codereview.chromium.org/599383002/diff/1/test/cctest/compiler/test-ma...
test/cctest/compiler/test-machine-operator-reducer.cc:168: if (tmp.IsSpecial()
&& !tmp.IsInfinite()) {
On 2014/09/25 09:08:41, titzer wrote:
> I would prefer to have this check in the callers.

Done.

[titzer, 2014-09-25 09:53:19]

lgtm

[vincent.belliard, 2014-09-29 08:34:29]

r 24270