Prototype CL for faster DOM accessors.

Prototype for faster DOM accessors.

- This is somewhat experimental; hence protected by #ifdef.
- This depends logically on crrev.com/1367953002, although it's
  technically independent.

R=epertoso@chromium.org, jochen@chromium.org
BUG=chromium:508898
LOG=N

[vogelheim, 2015-09-24 11:55:39]

vogelheim@chromium.org changed reviewers:
+ epertoso@chromium.org, jochen@chromium.org

[epertoso, 2015-09-25 08:22:54]

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc
File src/runtime/runtime-ptr.cc (right):

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:30: return *reinterpret_cast<aligned_ptr_t*>(&value);
just return reinterpret_cast<aligned_ptr>(value).

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:35: Object* object =
*reinterpret_cast<Object**>(&value);
As above, just reinterpret_cast<Object*>(value) is enough.

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:55: Object* object =
*reinterpret_cast<Object**>(&value);
And here too :).

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:64: HandleScope scope(isolate);
I think is that these functions should be UNREACHABLE, i.e. they are defined so
that turbofan can lower them.

OTOH, their implementation can be helpful to verify the correctness of the
function that's calling them. Should we have an #ifdef here?

[jochen (gone - plz use gerrit), 2015-09-25 11:20:12]

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc
File src/runtime/runtime-ptr.cc (right):

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:84: RUNTIME_FUNCTION(Runtime_PtrLoadOffsetInt) {
why not PtrTestIntAtOffset() or something?

[vogelheim, 2015-09-29 11:15:17]

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc
File src/runtime/runtime-ptr.cc (right):

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:30: return *reinterpret_cast<aligned_ptr_t*>(&value);
On 2015/09/25 08:22:53, epertoso wrote:
> just return reinterpret_cast<aligned_ptr>(value).

Done.

For some reason, I was dead sure that you could only reinterpret_cast<> between
pointer types. Apparently not. :)

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:35: Object* object =
*reinterpret_cast<Object**>(&value);
On 2015/09/25 08:22:53, epertoso wrote:
> As above, just reinterpret_cast<Object*>(value) is enough.

Done.

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:55: Object* object =
*reinterpret_cast<Object**>(&value);
On 2015/09/25 08:22:53, epertoso wrote:
> And here too :).

Done.

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:64: HandleScope scope(isolate);
On 2015/09/25 08:22:53, epertoso wrote:
> I think is that these functions should be UNREACHABLE, i.e. they are defined
so
> that turbofan can lower them.

Disagree. The reason why I want them (#ifdef-ed) in the code now - despite the
other parts not being in place - is that we can develop + experiment properly,
without having to patch the local client every time.

Long term you are probably right; but for the time being UNREACHABLE() strikes
me as the wrong thing to to.

> OTOH, their implementation can be helpful to verify the correctness of the
> function that's calling them. Should we have an #ifdef here?

Well... #ifdef V8_JS_ACCESSORS kinda does that. I think you mean to only #ifdef
the implementation of these runtime implementations; but there I disagree, as
above.

https://codereview.chromium.org/1292283005/diff/1/src/runtime/runtime-ptr.cc#...
src/runtime/runtime-ptr.cc:84: RUNTIME_FUNCTION(Runtime_PtrLoadOffsetInt) {
On 2015/09/25 11:20:12, jochen wrote:
> why not PtrTestIntAtOffset() or something?

Meanwhile, Enrico has done a good bit of work with this and he'll almost
certainly modify and extend these, once this CL is in. I'd prefer to let him
decide, depending on whether we really only need to mask bits, or possibly do
other things.

(An alternative would be for me to just drop this CL entirely, and let him check
in his version right away.)

[vogelheim, 2015-09-29 11:36:12]

vogelheim@chromium.org changed reviewers:
+ verwaest@chromium.org

[vogelheim, 2015-09-29 11:36:12]

@toon: Please let us know whether you're ok with this being checked in in this
early state.

[Toon Verwaest, 2015-09-30 13:57:19]

I don't have a particular beef with it.
Some minor comments.

https://codereview.chromium.org/1292283005/diff/20001/src/runtime/runtime-ptr.cc
File src/runtime/runtime-ptr.cc (right):

https://codereview.chromium.org/1292283005/diff/20001/src/runtime/runtime-ptr...
src/runtime/runtime-ptr.cc:39: static Object* fromAlignedPtr(aligned_ptr_t
value) {
Camelcase function names should start with uppercase.

https://codereview.chromium.org/1292283005/diff/20001/src/runtime/runtime-ptr...
src/runtime/runtime-ptr.cc:111: return (ptr == 0) ?
isolate->heap()->true_value()
return isolate->heap()->ToBoolean(ptr == nullptr);