Trace allocation of arrays

Trace allocation of arrays

BUG=
R=fschneider@google.com, srdjan@google.com

Committed: https://github.com/dart-lang/sdk/commit/e89b2eeaac0c43abf7d67e2d50fff842da0a4a57

[Cutch, 2015-07-10 02:05:43]

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

[Florian Schneider, 2015-07-10 13:02:39]

https://codereview.chromium.org/1226143011/diff/1/runtime/vm/stub_code_arm.cc
File runtime/vm/stub_code_arm.cc (right):

https://codereview.chromium.org/1226143011/diff/1/runtime/vm/stub_code_arm.cc...
runtime/vm/stub_code_arm.cc:640: const Class& cls =
Class::Handle(isolate->object_store()->array_class());
Can we make this isolate-independent code? I'm thinking ahead about precompiled
code, where we can't compile isolate-specific code.

Or, would it be easier to switch off the allocation tracing feature for
precompiled code?

[Cutch, 2015-07-10 13:16:29]

https://codereview.chromium.org/1226143011/diff/1/runtime/vm/stub_code_arm.cc
File runtime/vm/stub_code_arm.cc (right):

https://codereview.chromium.org/1226143011/diff/1/runtime/vm/stub_code_arm.cc...
runtime/vm/stub_code_arm.cc:640: const Class& cls =
Class::Handle(isolate->object_store()->array_class());
On 2015/07/10 13:02:39, Florian Schneider wrote:
> Can we make this isolate-independent code? I'm thinking ahead about
precompiled
> code, where we can't compile isolate-specific code.
> 

We can't without impacting performance. Also, all of the existing allocation
statistic collecting code is isolate specific and emitted from TryAllocate,
TryAllocateArray, and allocating stubs/intrinsics.

> Or, would it be easier to switch off the allocation tracing feature for
> precompiled code?

Should be easy enough.

[Florian Schneider, 2015-07-10 13:25:23]

lgtm

https://codereview.chromium.org/1226143011/diff/1/runtime/vm/stub_code_arm.cc
File runtime/vm/stub_code_arm.cc (right):

https://codereview.chromium.org/1226143011/diff/1/runtime/vm/stub_code_arm.cc...
runtime/vm/stub_code_arm.cc:640: const Class& cls =
Class::Handle(isolate->object_store()->array_class());
On 2015/07/10 13:16:29, Cutch wrote:
> On 2015/07/10 13:02:39, Florian Schneider wrote:
> > Can we make this isolate-independent code? I'm thinking ahead about
> precompiled
> > code, where we can't compile isolate-specific code.
> > 
> 
> We can't without impacting performance. Also, all of the existing allocation
> statistic collecting code is isolate specific and emitted from TryAllocate,
> TryAllocateArray, and allocating stubs/intrinsics.
> 
> > Or, would it be easier to switch off the allocation tracing feature for
> > precompiled code?
> 
> Should be easy enough.

Yes, also the existing allocation code has isolate-specific addresses embedded
for performance. We may need a completely separate code path for precompiled
code then.

[koda, 2015-07-10 14:52:03]

On 2015/07/10 13:25:23, Florian Schneider wrote:
> lgtm
> 
> https://codereview.chromium.org/1226143011/diff/1/runtime/vm/stub_code_arm.cc
> File runtime/vm/stub_code_arm.cc (right):
> 
>
https://codereview.chromium.org/1226143011/diff/1/runtime/vm/stub_code_arm.cc...
> runtime/vm/stub_code_arm.cc:640: const Class& cls =
> Class::Handle(isolate->object_store()->array_class());
> On 2015/07/10 13:16:29, Cutch wrote:
> > On 2015/07/10 13:02:39, Florian Schneider wrote:
> > > Can we make this isolate-independent code? I'm thinking ahead about
> > precompiled
> > > code, where we can't compile isolate-specific code.
> > > 
> > 
> > We can't without impacting performance. Also, all of the existing allocation
> > statistic collecting code is isolate specific and emitted from TryAllocate,
> > TryAllocateArray, and allocating stubs/intrinsics.
> > 
> > > Or, would it be easier to switch off the allocation tracing feature for
> > > precompiled code?
> > 
> > Should be easy enough.
> 
> Yes, also the existing allocation code has isolate-specific addresses embedded
> for performance. We may need a completely separate code path for precompiled
> code then.

We could use thread-local recording of allocation stats, and then accumulate
them in each isolate. The THR register works fine for precompiled code.

[srdjan, 2015-07-10 16:36:57]

lgtm

[Cutch, 2015-07-10 20:06:23]

Committed patchset #3 (id:40001) manually as
e89b2eeaac0c43abf7d67e2d50fff842da0a4a57 (presubmit successful).