Include all the code in code creation log events. The code object header size...

Include all the code in code creation log events. The code object header size is now added to the size Before ticks the last instructions of a JavaScript functions ended up as unaccounted in the profile.Include ticks in the the last created code object in the profile.Show the unaccounted ticks in the profile as a percentage together with the rest. Added an option to ignore unaccounted ticks in the percentage calculation.

Committed: http://code.google.com/p/v8/source/detail?r=1295

[Mads Ager (chromium), 2009-02-18 08:31:28]

STV!

Looking forward to having less unaccounted ticks!  I noticed yesterday that I
had a pretty large number of unaccounted ticks in my profiles - this change will
probably fix that.  :-)

[Mikhail Naganov, 2009-02-18 08:36:25]

LGTM

I see no problems here, except nits.

http://codereview.chromium.org/21410/diff/1/5
File src/log.cc (right):

http://codereview.chromium.org/21410/diff/1/5#newcode676
Line 676: code->instruction_size() + Code::kHeaderSize);
nit: maybe extract a method here, as code gets repeated

http://codereview.chromium.org/21410/diff/1/2
File tools/tickprocessor.py (right):

http://codereview.chromium.org/21410/diff/1/2#newcode153
Line 153: # Falg indicating whether to ignore unaccounted ticks in the report
typo: Flag

[iposva, 2009-02-18 09:08:41]

Please consider using the proper accessors to get code start as outlined below.

Thanks,
-Ivan

http://codereview.chromium.org/21410/diff/1/5
File src/log.cc (right):

http://codereview.chromium.org/21410/diff/1/5#newcode676
Line 676: code->instruction_size() + Code::kHeaderSize);
This would be much simpler if you used code->instruction_start() and
code->instruction_size() instead of doing the math here. In particular if ever
the Code object changes its internal layout your calculation here is off.

This will also address Mikhail's comment.

[Søren Thygesen Gjesse, 2009-02-18 10:24:10]

http://codereview.chromium.org/21410/diff/1/5
File src/log.cc (right):

http://codereview.chromium.org/21410/diff/1/5#newcode676
Line 676: code->instruction_size() + Code::kHeaderSize);
On 2009/02/18 09:08:41, iposva wrote:
> This would be much simpler if you used code->instruction_start() and
> code->instruction_size() instead of doing the math here. In particular if ever
> the Code object changes its internal layout your calculation here is off.
> 
> This will also address Mikhail's comment.

Initially I went for instruction_start for all code events. However code-move
will report the object address, and we would have to do math there instead.

Added a function for doing the calculation with an assert for checking the
assumptions about the layout of the code object.

One need to run profiling in a debug build to hit the assertion though.

http://codereview.chromium.org/21410/diff/1/2
File tools/tickprocessor.py (right):

http://codereview.chromium.org/21410/diff/1/2#newcode153
Line 153: # Falg indicating whether to ignore unaccounted ticks in the report
On 2009/02/18 08:36:25, Mikhail Naganov wrote:
> typo: Flag

Done.