Print reason for disabling optimization. Kill --trace-bailout flag.

Print reason for disabling optimization. Kill --trace-bailout flag.

The reason for disabling optimization of a given function is carried around in
CompilationInfo. The new mechanism is general enough that --trace-opt now
subsumes everything --trace-bailout could print, so we nuked the latter flag.

Committed: https://code.google.com/p/v8/source/detail?r=12391

[Jakob Kummerow, 2012-08-27 14:26:02]

LGTM with an idea for further simplification.

http://codereview.chromium.org/10868106/diff/4001/src/lithium.cc
File src/lithium.cc (right):

http://codereview.chromium.org/10868106/diff/4001/src/lithium.cc#newcode393
src/lithium.cc:393: LChunk* LChunk::NewChunk(HGraph* graph, CompilationInfo*
info) {
Instead of passing in |info|, can't you just use graph->info() (as line 403 does
anyway)?

[Vyacheslav Egorov (Google), 2012-08-27 14:33:43]

qq: does it correctly attribute bailout that happened inside one of the function
we are inlining to it or does it accidentally attribute it to the caller?

[Sven Panne, 2012-08-28 07:17:07]

Regarding inlined functions: Crankshaft keeps a stack of FunctionStates with an
entry for each inlining level, and each state has its own CompilationInfo.
Therefore we don't attribute things to the wrong level, but I think we might
lose some info when popping the stack. But I have to admit that I can't come up
with a good repro. :-)

Landing...

http://codereview.chromium.org/10868106/diff/4001/src/lithium.cc
File src/lithium.cc (right):

http://codereview.chromium.org/10868106/diff/4001/src/lithium.cc#newcode393
src/lithium.cc:393: LChunk* LChunk::NewChunk(HGraph* graph, CompilationInfo*
info) {
On 2012/08/27 14:26:02, Jakob wrote:
> Instead of passing in |info|, can't you just use graph->info() (as line 403
does
> anyway)?

Good point, I missed that one after refactoring back and forth. Not needing
another argument is a good sign that CompilationInfo is indeed the right place
to keep the bailout reason. Done.