Don't bother trying to reuse object pool entries for ICs, type test caches, or edge counters as the…

Don't bother trying to reuse object pool entries for ICs, type test caches, or edge counters as they will have unique entries anyway.

Avoids lookups of object pool entries, which dominates in very large functions such as regex.

First program from http://dartbug.com/23249 goes from 66.439s to 10.918s on X64 (6.1x).

R=srdjan@google.com

Committed: https://github.com/dart-lang/sdk/commit/77c206f2318260d109944567918280e9fbef0dde

[rmacnak, 2015-07-07 21:16:16]

rmacnak@google.com changed reviewers:
+ fschneider@google.com, srdjan@google.com

[srdjan, 2015-07-07 21:57:24]

https://codereview.chromium.org/1231463002/diff/20001/runtime/vm/assembler_ar...
File runtime/vm/assembler_arm.cc (right):

https://codereview.chromium.org/1231463002/diff/20001/runtime/vm/assembler_ar...
runtime/vm/assembler_arm.cc:1601: }
You could create a third function (private):
LoadObjectHelper(Register rd, const Object& object, Condition cond, bool
is_unique) {
  ...
  const int32_t offset = is_unique ?
      ObjectPool::element_offset(object_pool_wrapper_.AddObject(object)) :
      ObjectPool::element_offset(object_pool_wrapper_.FindObject(object));
  ...
}

Or you could use existing LoadObject and add as last argument 'bool is_unique =
false'.

What do you think?

[rmacnak, 2015-07-07 23:10:59]

https://codereview.chromium.org/1231463002/diff/20001/runtime/vm/assembler_ar...
File runtime/vm/assembler_arm.cc (right):

https://codereview.chromium.org/1231463002/diff/20001/runtime/vm/assembler_ar...
runtime/vm/assembler_arm.cc:1601: }
On 2015/07/07 21:57:23, srdjan wrote:
> You could create a third function (private):
> LoadObjectHelper(Register rd, const Object& object, Condition cond, bool
> is_unique) {
>   ...
>   const int32_t offset = is_unique ?
>       ObjectPool::element_offset(object_pool_wrapper_.AddObject(object)) :
>       ObjectPool::element_offset(object_pool_wrapper_.FindObject(object));
>   ...
> }
> 
> Or you could use existing LoadObject and add as last argument 'bool is_unique
=
> false'.
> 
> What do you think?

I prefer a helper. Did that.

[srdjan, 2015-07-07 23:21:22]

lgtm

[rmacnak, 2015-07-07 23:40:33]

Committed patchset #3 (id:40001) manually as
77c206f2318260d109944567918280e9fbef0dde (presubmit successful).

[Florian Schneider, 2015-07-08 08:29:57]

Those objects are probably unique, but wouldn't adding a proper hashcode
function to ObjIndexPair solve the problem already?