Introduce cache of resolved names in library

Introduce cache of resolved names in library

A name that is not resolved in the local scope gets looked up
in the global scope of the library to which the code being compiled
belongs. The local name dictionary gets searched first, then the
dictionary of each imported library (and and their re-exported
libraries.) Each dictionary lookup includes the lookup of the
mangled getter and setter names, which means that we concatenate
the same name many times.

This change introduces a cache that stores the result of a previous
name lookup, including negative lookup results. The latter is important
to speed up the resolution of names that are not in the global name
space, e.g. class members like .length in lists.

Experiments running dart2js (with VM option --compile_all) show
that this reduces the number of name mangling calls by a factor of 10
and speeds up compiling dart2js by 15%.

R=asiva@google.com

Committed: https://code.google.com/p/dart/source/detail?r=32385

[siva, 2014-02-06 21:01:29]

lgtm

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc
File runtime/vm/object.cc (right):

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc#newcode...
runtime/vm/object.cc:7915: AddResolvedName(name, obj);
AddToResolvedNamesCache would be a more readable method name here.

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc#newcode...
runtime/vm/object.cc:7977: // The entry cache[N-1] is used a counter
used as a counter

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc#newcode...
runtime/vm/object.cc:8160: InvalidateResolvedName(class_name);
Should we also invalidate this cache when a class gets compiled lazily.

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc#newcode...
runtime/vm/object.cc:8577: const intptr_t initial_name_cache_size = 64;
const intptr_t kInitialNameCacheSize = 64;

[hausner, 2014-02-06 21:31:45]

Thank you.

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc
File runtime/vm/object.cc (right):

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc#newcode...
runtime/vm/object.cc:7915: AddResolvedName(name, obj);
On 2014/02/06 21:01:29, siva wrote:
> AddToResolvedNamesCache would be a more readable method name here.

Done.

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc#newcode...
runtime/vm/object.cc:7977: // The entry cache[N-1] is used a counter
On 2014/02/06 21:01:29, siva wrote:
> used as a counter

Done

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc#newcode...
runtime/vm/object.cc:8160: InvalidateResolvedName(class_name);
On 2014/02/06 21:01:29, siva wrote:
> Should we also invalidate this cache when a class gets compiled lazily.

I don't think this is necessary. The name of the class is in the library scope
already when it gets compiled, is it not? And by compiling the class nothing
gets added to the library's name space, only to the name space of the class.

https://codereview.chromium.org/135123011/diff/1/runtime/vm/object.cc#newcode...
runtime/vm/object.cc:8577: const intptr_t initial_name_cache_size = 64;
On 2014/02/06 21:01:29, siva wrote:
> const intptr_t kInitialNameCacheSize = 64;

Done.

[hausner, 2014-02-06 21:32:37]

Committed patchset #2 manually as r32385 (presubmit successful).

[rmacnak, 2014-02-07 18:44:03]

On 2014/02/06 21:32:37, hausner wrote:
> Committed patchset #2 manually as r32385 (presubmit successful).

Sped up cold access to the mirror system by 50% too!