[PATCH] Indirect call target profiling related profile reader/writer changes

[Ivan Baev]

>> Ok, let's start from the basics.  For profile guided devirtualization,
>> you're constructing a cache from (something) to function pointer and
>> using that cache lookup to enable inlining of the hot target.  You have
>> two standard choices on what to use as your cache key: the result of the
>> virtual lookup and the inputs to the virtual lookup.
>>
>> Option 1 - Inputs to virtual lookup
>> if ((receiver, vtable index) == what I predicted)
>>    tartget_I_predicted(); // inline me!!
>> else {
>>    target = full virtual dispatch();
>>    target();
>> }
>>
>> Option 2 - result of virtual lookup
>> target = full virtual dispatch();
>> if ('target' == what I predicted)
>>    tartget_I_predicted(); // inline me!!
>> else {
>>    target();
>> }
>>
>> You seem to be proposing option 2.  I'm saying that I'm used to seeing
>> option 1 used.  Both approaches have their appeal, I'm just asking you
>> to explain *why* you've chosen the one you apparently have.
>
> Not all indirect calls occur from C++ like codes. We're profiling and
> optimizing out indirect calls from C codes as well. We're seeing up to 8%
> gains on individual benchmarks in spec. This was measured on our platform.
>

Our initial motivation was to handle standard indirect calls, that is,
 if ('target' == what I predicted)
    target_I_predicted(); // inline me!!
 else {
    target();
 }

At the start of the indirect call promotion pass we see a virtual call as
an indirect call so we can say we handle virtual calls through option 2.
 target = full virtual dispatch();
 if ('target' == what I predicted)
    target_I_predicted(); // inline me!!
 else {
    target();
 }

Now assuming that full virtual dispatch() is something like
  { vtable = load [receiver + #VToffset];
    target = load [vtable + index]; }

under certain conditions we might be able to enhance option 2 above to
  vtable =  load [receiver + #VToffset];
  if (vtable == vtable_of_what I predicted)
    target_I_predicted(); // inline me!!
  else {
    target = load [vtable + index];
    target();
  }

Ivan