Simplify the transitions in the Binary Op ICs. Now a single call...

Simplify the transitions in the Binary Op ICs.  Now a single call
to the runtime will both patch in the more specialized binary op
stub and calculate the answer.  This eliminates the need to call
both the rest of the binary op and the patching runtime call.  The
runtime routines are altered to be more agressive in returning
Smis so we don't get spurious heap numbers as inputs to binary ops
while we are patching the binary op ICs.

Committed: http://code.google.com/p/v8/source/detail?r=5026

[Kasper Lund, 2010-07-06 11:56:35]

LGTM. As we discussed, it would be nice to start moving this tricky code out of
the codegen files and add some abstractions that makes it easier to generate
(even) more specialized stub variants.

http://codereview.chromium.org/2843049/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/2843049/diff/1/2#newcode7838
src/arm/codegen-arm.cc:7838: if (generate_code_to_calculate_answer ||
slow_reverse.is_linked()) {
Maybe just move the if (lhs.is(r0)) code into the next then-part that's guarded
by if (generate_code...)?

http://codereview.chromium.org/2843049/diff/1/2#newcode7849
src/arm/codegen-arm.cc:7849: if (generate_code_to_calculate_answer ||
slow.is_linked()) {
I would consider negating the condition and returning at this point.

http://codereview.chromium.org/2843049/diff/1/3
File src/ia32/codegen-ia32.cc (right):

http://codereview.chromium.org/2843049/diff/1/3#newcode10233
src/ia32/codegen-ia32.cc:10233: // Patch the caller to an appropriate
specialized stub and return the
I guess it doesn't really return anymore? Does it?

http://codereview.chromium.org/2843049/diff/1/5
File src/runtime.cc (right):

http://codereview.chromium.org/2843049/diff/1/5#newcode5580
src/runtime.cc:5580: return Heap::NumberFromDouble(x + y);
No more AllocateHeapNumber left in runtime.cc?

http://codereview.chromium.org/2843049/diff/1/8
File test/mjsunit/value-of.js (right):

http://codereview.chromium.org/2843049/diff/1/8#newcode33
test/mjsunit/value-of.js:33: assertThrows(function() {o + 1; }, MyException);
{o + 1; } => { o + 1 }

[Erik Corry, 2010-07-06 12:53:21]

http://codereview.chromium.org/2843049/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/2843049/diff/1/2#newcode7838
src/arm/codegen-arm.cc:7838: if (generate_code_to_calculate_answer ||
slow_reverse.is_linked()) {
On 2010/07/06 11:56:35, Kasper Lund wrote:
> Maybe just move the if (lhs.is(r0)) code into the next then-part that's
guarded
> by if (generate_code...)?

Done.

http://codereview.chromium.org/2843049/diff/1/2#newcode7849
src/arm/codegen-arm.cc:7849: if (generate_code_to_calculate_answer ||
slow.is_linked()) {
On 2010/07/06 11:56:35, Kasper Lund wrote:
> I would consider negating the condition and returning at this point.

Done.

http://codereview.chromium.org/2843049/diff/1/3
File src/ia32/codegen-ia32.cc (right):

http://codereview.chromium.org/2843049/diff/1/3#newcode10233
src/ia32/codegen-ia32.cc:10233: // Patch the caller to an appropriate
specialized stub and return the
On 2010/07/06 11:56:35, Kasper Lund wrote:
> I guess it doesn't really return anymore? Does it?

Well it's a tail call so it returns the answer to the caller of the stub.

http://codereview.chromium.org/2843049/diff/1/5
File src/runtime.cc (right):

http://codereview.chromium.org/2843049/diff/1/5#newcode5580
src/runtime.cc:5580: return Heap::NumberFromDouble(x + y);
On 2010/07/06 11:56:35, Kasper Lund wrote:
> No more AllocateHeapNumber left in runtime.cc?

I removed one for integer Math.pow, but the rest were very unlikely to be Smis
so I left them as plain AllocateHeapNumber because it is faster for things that
are not Smis.

http://codereview.chromium.org/2843049/diff/1/8
File test/mjsunit/value-of.js (right):

http://codereview.chromium.org/2843049/diff/1/8#newcode33
test/mjsunit/value-of.js:33: assertThrows(function() {o + 1; }, MyException);
On 2010/07/06 11:56:35, Kasper Lund wrote:
> {o + 1; } => { o + 1 }

Done.