Let NaN flow as double into HBranch

Let NaN flow as double into HBranch + some minor improvements

BUG=
R=verwaest@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=15246

[Toon Verwaest, 2013-06-18 17:01:52]

first round of comments.

https://codereview.chromium.org/17082003/diff/3001/src/hydrogen-instructions.cc
File src/hydrogen-instructions.cc (right):

https://codereview.chromium.org/17082003/diff/3001/src/hydrogen-instructions....
src/hydrogen-instructions.cc:1290: if
(expected_input_types_.ContainsAnyOf(number_types)) {
Are you sure you want to make smi+undefined double rather than tagged? Seems
like tagged would be faster; given that smis are valid tagged objects; and you
need to do a smi-check anyway.

I'd rather make undefined a special case of Contains(HEAP_NUMBER).

https://codereview.chromium.org/17082003/diff/3001/src/ia32/lithium-codegen-i...
File src/ia32/lithium-codegen-ia32.cc (right):

https://codereview.chromium.org/17082003/diff/3001/src/ia32/lithium-codegen-i...
src/ia32/lithium-codegen-ia32.cc:2139: } else if (type.IsHeapNumber()) {
ASSERT(!info()->IsStub());
XMM is not available for stubs.

https://codereview.chromium.org/17082003/diff/3001/src/x64/lithium-x64.cc
File src/x64/lithium-x64.cc (right):

https://codereview.chromium.org/17082003/diff/3001/src/x64/lithium-x64.cc#new...
src/x64/lithium-x64.cc:1022: expected != ToBooleanStub::all_types()) {
Can you add a comment what expected == all_types means?
Also, can we make ia32/x64 and ARM agree on &&,! vs ||?

[oliv, 2013-06-19 10:06:31]

PTAL, sry for rebase, forgot to store original branch point...

On 2013/06/18 17:01:52, Toon Verwaest wrote:
>
https://codereview.chromium.org/17082003/diff/3001/src/hydrogen-instructions....
> src/hydrogen-instructions.cc:1290: if
> (expected_input_types_.ContainsAnyOf(number_types)) {
> Are you sure you want to make smi+undefined double rather than tagged? Seems
> like tagged would be faster; given that smis are valid tagged objects; and you
> need to do a smi-check anyway.
> 
> I'd rather make undefined a special case of Contains(HEAP_NUMBER).

agree

>
https://codereview.chromium.org/17082003/diff/3001/src/ia32/lithium-codegen-i...
> src/ia32/lithium-codegen-ia32.cc:2139: } else if (type.IsHeapNumber()) {
> ASSERT(!info()->IsStub());
> XMM is not available for stubs.

ok

>
https://codereview.chromium.org/17082003/diff/3001/src/x64/lithium-x64.cc#new...
> src/x64/lithium-x64.cc:1022: expected != ToBooleanStub::all_types()) {
> Can you add a comment what expected == all_types means?
> Also, can we make ia32/x64 and ARM agree on &&,! vs ||?

tried to make it more self explaining

[Toon Verwaest, 2013-06-20 12:45:59]

lgtm with a nit

https://codereview.chromium.org/17082003/diff/13001/src/arm/lithium-arm.cc
File src/arm/lithium-arm.cc (right):

https://codereview.chromium.org/17082003/diff/13001/src/arm/lithium-arm.cc#ne...
src/arm/lithium-arm.cc:1002: // deoptimization environment. Except in the
Generic case.
If the instruction is generic no environment is needed since all cases are
handled.

[oliv, 2013-06-20 13:51:14]

Committed patchset #4 manually as r15246 (presubmit successful).