Lots of web pages have really frequently firing timers that keep the
profiler thread spinning if we require a period of JS inactivity
before suspending the profiler. While it's possible to throttle it by
increasing the sleep delay and adjusting the duration of the required
inactive period, it seemed much simpler to just stop it immediately on
exiting JS.
Stopping the profiler this way effectively turned off two optimization
heuristics: 1) eager optimization (it's reset on waking up the
profiler and now the profiler wakes up much more frequently) and 2)
optimization throttling based on JS to non-JS state ratio (the ratio
is now 100%). I removed these two heuristics and found no performance
regressions so far.
R=ager@chromium.org
BUG=crbug.com/77625
TEST=none
Review URL: http://codereview.chromium.org/7274024
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8472 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Optimized frames are now handled by the debugger. When discovering optimized frames during stack inspection in the debugger they are "deoptimized" using the normal deoptimization code and the deoptimizer output information is used to provide frame information to the debugger.
Before this change the debugger reported each optimized frame as one frame no matter the number of inlined functuions that might have been called inside of it. Also all locals where reported as undefined. Locals can still be reposted as undefined when their value is not "known" by the optimized frame.
As the structures used to calculate the output frames when deoptimizing are not GC safe the information for the debugger is copied to another structure (DeoptimizedFrameInfo) which is registered with the global deoptimizer data and processed during GC.
R=fschneider@chromium.org
BUG=v8:1140
TEST=test/mjsunit/debug-evaluate-locals-optimized*
Review URL: http://codereview.chromium.org//7230045
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8464 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Only IA32 version for now. I'll start porting.
Strict mode functions are to get 'undefined' as the receiver when
called with an implicit receiver. Modes are bad! It forces us to have
checks on all function calls.
This change attempts to limit the cost by passing information about
whether or not a call is with an implicit or explicit receiver in ecx
as part of the calling convention. The cost is setting ecx on all
calls and checking ecx on entry to strict mode functions.
Implicit/explicit receiver state has to be maintained by ICs. Various
stubs have to not clobber ecx or save and restore it.
CallFunction stub needs to check if the receiver is implicit when it
doesn't know from the context.
Review URL: http://codereview.chromium.org/7039036
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8040 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
- mutual inlining strict and non-strict functions in crankshaft.
- assignment to undefined variable with eval in scope.
- propagation of strict mode through lazy compilation.
BUG=
TEST=test/mjsunit/strict-mode.js test/mjsunit/strict-mode-opt.js
Review URL: http://codereview.chromium.org/6814012
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7561 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
While trying to fix Mac and Windows versions for this change:
http://codereview.chromium.org/6771047/, I figured out, that we
already store an isolate in StackFrameIterator, so we can use it in
frame objects, instead of requiring it from caller.
I've changed iterators usage to the following scheme: whenever a
caller maintains an isolate pointer, it just passes it to stack
iterator, and no more worries about passing it to frame content
accessors. If a caller uses current isolate, it can omit passing it
to iterator, in this case, an iterator will use the current isolate,
too.
There was a special case with LiveEdit, which creates
detached copies of frame objects.
R=vitalyr@chromium.org
BUG=none
TEST=none
Review URL: http://codereview.chromium.org/6794019
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7499 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Crankshaft is now the default on all platforms. This is the first
patch on the way to removing the classic code generator from the
system.
This time with no removal of the crankshaft flag. --nocrankshaft is
not at all the same as --always-full-compiler which I had used instead
for testing. That was what caused timeouts on the buildbots because of
repeated attempts to optimize hot functions. It makes sense to keep
the crankshaft flag in case you want to run only with the full
compiler and with no adaptive compilation.
R=vitalyr@chromium.org
Review URL: http://codereview.chromium.org/6759070
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7486 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Because we might throw away code when doing code flushing we need to
set the optimizable flag to false in CompileLaze if this has been set
on the shared function info. This is the only place where this can
happen, since we always exchange the code with the laze compile stub
when doing code flushing.
The comment in AbortAndDisable actually states that this is already
the case (and that comment should now be ok).
Review URL: http://codereview.chromium.org/6685044
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7378 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
Change the way we construct the graph for polymorphic loads to match that of
polymorphic stores.
Introduce a stack-allocated helper for saving and restoring all the
function-specific graph builder state that needs to change when we begin
translating an inlined function. Make this class authoritative by moving
redundant state out of the builder and deferring to the current function's
state.
Ensure that we always print a tracing message when abandoning an inlining
attempt.
Review URL: http://codereview.chromium.org/6628012
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@7074 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
The main issue was due to multiple recompilations of functions. Now
code objects are grouped by function using SFI object address.
JSFunction objects are no longer tracked, instead we track SFI object
moves. To pick a correct code version, we now sample return addresses
instead of JSFunction addresses.
tools/{linux|mac|windows}-tickprocessor scripts differentiate
between code optimization states for the same function
(using * and ~ prefixes introduced earlier).
DevTools CPU profiler treats all variants of function code as
a single function.
ll_prof treats each optimized variant as a separate entry, because
it can disassemble each one of them.
tickprocessor.py not updated -- it is deprecated and will be removed.
BUG=v8/1087,b/3178160
TEST=all existing tests pass, including Chromium layout tests
Review URL: http://codereview.chromium.org/6551011
git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@6902 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
