Make deoptimization stress count global.

Make deoptimization stress count global.

Store the deopt stress counter per isolate instead of per shared function info. The old field is removed.

Enable output of the counter value with a new flag.

R=mstarzinger@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=15739

[Michael Achenbach, 2013-07-16 14:23:44]

PTAL

[danno, 2013-07-16 14:47:00]

https://codereview.chromium.org/19383002/diff/1/src/ia32/lithium-codegen-ia32.cc
File src/ia32/lithium-codegen-ia32.cc (right):

https://codereview.chromium.org/19383002/diff/1/src/ia32/lithium-codegen-ia32...
src/ia32/lithium-codegen-ia32.cc:935: ExternalReference count =
ExternalReference::stress_deopt_count(isolate());
For extra points: Put this code in a code stub with the bulk of this code that
returns whether to deopt or not in a condition code flag. That stub can also
call to a piece of runtime C++ in the case that a deopt will happen an print out
a message if the trace flag is true.

[Michael Starzinger, 2013-07-16 15:23:35]

LGTM (if comments are addressed).

https://codereview.chromium.org/19383002/diff/1/src/flag-definitions.h
File src/flag-definitions.h (right):

https://codereview.chromium.org/19383002/diff/1/src/flag-definitions.h#newcod...
src/flag-definitions.h:256: DEFINE_bool(trace_deopt_stress, false, "trace number
of possible deopt points")
This doesn't really trace anything but just print statistics at tear down. Can
we s/trace_deopt_stress/print_deopt_stress/ or similar?

https://codereview.chromium.org/19383002/diff/1/src/isolate.cc
File src/isolate.cc (right):

https://codereview.chromium.org/19383002/diff/1/src/isolate.cc#newcode1903
src/isolate.cc:1903: PrintF(stdout, "=== Stress deopt counter: %u\n",
stress_deopt_count_);
nit: Let's drop the '===' marker in the output.

[Michael Achenbach, 2013-07-17 16:05:21]

PTAL at patch 2.

- Changed flag name to "print"
- Removed SMI conversion to make it actually work
- Stubs will follow in a different CL
- Lines with the === prefix are omitted in test suites that compare output 1:1 -
keeping this prefix means the tests could run and be analyzed in the same time

[Michael Starzinger, 2013-07-17 17:08:38]

Still LGTM from my end.

[Michael Achenbach, 2013-07-18 08:12:08]

Committed patchset #3 manually as r15739 (presubmit successful).