Count ICs that have type information.

Count ICs that have type information.

Hidden behind --type-info-threshold=X flag, usage dependent on experimental profiler.

Committed: https://code.google.com/p/v8/source/detail?r=10753

[Jakob Kummerow, 2012-02-15 11:09:17]

Please take a look.

[Vyacheslav Egorov (Chromium), 2012-02-15 12:07:39]

It is intentional that we only grow counter and never decrease it (even when we
clear ICs or even discard code?)

Also strongly recommend to move these counters into the code object (that would
also simplify incrementing/decrementing logic).

We already have array that handles ast_id -> function target there. I sense that
that array can be turned into generic host for type feedback information
collected by the code.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ia32/full-codegen-i...
File src/ia32/full-codegen-ia32.cc (right):

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ia32/full-codegen-i...
src/ia32/full-codegen-ia32.cc:4298: info_->increment_number_of_ics();
Please wrap increment and __ call together into a helper function called
CallIC() instead of spreading it.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ic.cc
File src/ic.cc (right):

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ic.cc#newcode335
src/ic.cc:335: if (address > code->instruction_start() &&
I think Code has Contains method.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ic.cc#newcode337
src/ic.cc:337: shared->set_ic_typeinfo_count(shared->ic_typeinfo_count() + 1);
Please add a small comment that this what this code actually does: it tries to
find s.f.i that owns code that contains given IC.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/runtime-profiler.cc
File src/runtime-profiler.cc (right):

https://chromiumcodereview.appspot.com/9403009/diff/1/src/runtime-profiler.cc...
src/runtime-profiler.cc:114: PrintF(", ICs with typeinfo: %d/%d (%d%%)",
I sense SIGFPE here for code objects without ICs.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/runtime-profiler.cc...
src/runtime-profiler.cc:265: int ic_typeinfo =
function->shared()->ic_typeinfo_count();
I would rename ic_typeinfo_count into ic_with_typeinfo_count

https://chromiumcodereview.appspot.com/9403009/diff/1/src/runtime-profiler.cc...
src/runtime-profiler.cc:267: int percentage = 100 * ic_typeinfo / ic_total;
I sense SIGFPE here for code objects without ICs.

[Jakob Kummerow, 2012-02-17 16:07:52]

I have uploaded a new patch set that addresses your comments and includes your
suggestions. Please take another look.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ia32/full-codegen-i...
File src/ia32/full-codegen-ia32.cc (right):

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ia32/full-codegen-i...
src/ia32/full-codegen-ia32.cc:4298: info_->increment_number_of_ics();
On 2012/02/15 12:07:39, Vyacheslav Egorov wrote:
> Please wrap increment and __ call together into a helper function called
> CallIC() instead of spreading it.

Done.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ic.cc
File src/ic.cc (right):

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ic.cc#newcode335
src/ic.cc:335: if (address > code->instruction_start() &&
On 2012/02/15 12:07:39, Vyacheslav Egorov wrote:
> I think Code has Contains method.

Thanks for the suggestion; but this is obsolete anyway due to your other
suggestions :-)

https://chromiumcodereview.appspot.com/9403009/diff/1/src/ic.cc#newcode337
src/ic.cc:337: shared->set_ic_typeinfo_count(shared->ic_typeinfo_count() + 1);
On 2012/02/15 12:07:39, Vyacheslav Egorov wrote:
> Please add a small comment that this what this code actually does: it tries to
> find s.f.i that owns code that contains given IC.
 
This is obsolete too.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/runtime-profiler.cc
File src/runtime-profiler.cc (right):

https://chromiumcodereview.appspot.com/9403009/diff/1/src/runtime-profiler.cc...
src/runtime-profiler.cc:114: PrintF(", ICs with typeinfo: %d/%d (%d%%)",
On 2012/02/15 12:07:39, Vyacheslav Egorov wrote:
> I sense SIGFPE here for code objects without ICs.

Done.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/runtime-profiler.cc...
src/runtime-profiler.cc:265: int ic_typeinfo =
function->shared()->ic_typeinfo_count();
On 2012/02/15 12:07:39, Vyacheslav Egorov wrote:
> I would rename ic_typeinfo_count into ic_with_typeinfo_count

Done.

https://chromiumcodereview.appspot.com/9403009/diff/1/src/runtime-profiler.cc...
src/runtime-profiler.cc:267: int percentage = 100 * ic_typeinfo / ic_total;
On 2012/02/15 12:07:39, Vyacheslav Egorov wrote:
> I sense SIGFPE here for code objects without ICs.

Done.

[Vyacheslav Egorov (Chromium), 2012-02-20 08:32:13]

lgtm

https://chromiumcodereview.appspot.com/9403009/diff/10001/src/objects-inl.h
File src/objects-inl.h (right):

https://chromiumcodereview.appspot.com/9403009/diff/10001/src/objects-inl.h#n...
src/objects-inl.h:3627: 
accidental edit?