Closed Bug 512744 Opened 15 years ago Closed 15 years ago

code size measurement in fragment profiler is bogus

Categories

(Core :: JavaScript Engine, defect)

x86
Linux
defect
Not set
normal

Tracking

()

RESOLVED FIXED

People

(Reporter: jseward, Assigned: graydon)

References

Details

Attachments

(1 file)

w.r.t the fragprofiling patch in bug 503424, the mechanism for measuring the static code size (the amount of generated code) is completely bogus now that CodeAlloc has been introduced. It would be really useful to separately measure main-trace (hot) vs side-exit (cold) insn byte counts. From my poking around in Assembler.cpp I think this should be pretty straightforward.
Attached patch implement it correctly (deleted) — Splinter Review
Julian: this patch implements code-size (and exit-size) counting correctly, as far as I can tell. The numbers look reasonable anyway. It sits on your most recent patch to bug 503424. Feel free to just integrate it into that bug. I switched the byte counters to size_t; not sure if this is ideal given that you're truncating to 5 digits on output. I think I'd prefer to use appropriate sizes up-front, until the reporting line, rather than having the formatting choice propagate to the types all through the code.
Er, oops, the comment hunk of line 669 of Assembler.cpp is a stale and no-longer-meaningful comment, don't copy that bit.
(In reply to comment #1) > Julian: this patch implements code-size (and exit-size) counting correctly, Graydon, thanks v. much. Will assimilate this into the bug 503424 patch.
This has been fixed by merging the patch in comment #1 into the patch for bug 503424.
Status: NEW → RESOLVED
Closed: 15 years ago
Resolution: --- → FIXED
Flags: in-testsuite-
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: