Added a EvalCache that caches eval'ed scripts and compiled function boilerpla...

Added a EvalCache that caches eval'ed scripts and compiled function boilerplate.
The cache is a hashtable that takes String as key and JSFunction as the value.

Caches are cleared before mark-compact GC's.

Currently I don't put caps on cache size, string size, etc.

This cuts date-parse-totfe.js runtime by half.



Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=173

[Feng Qian, 2008-09-05 00:13:06]

My old client with this change was lost due to svn update.

I decide to use gcl to recreate this CB.

[Kasper Lund, 2008-09-05 09:16:29]

This looks great. I think you should refactor it a bit and introduce a new class
- CompilationCache - that deals with the different tables instead of just
expanding the Heap class to deal with it.

Feel free to submit it as it is and do the refactoring in another CL. I'm
looking forward to seeing the improvements this may bring us for scripts too.

http://codereview.chromium.org/457/diff/1/8
File src/heap.h (right):

http://codereview.chromium.org/457/diff/1/8#newcode538
Line 538: // EvalCache caches function boilerplates for compiled scripts
I would move this code out into a separate class with static operations;
something like CompilationCache maybe?

http://codereview.chromium.org/457/diff/1/8#newcode544
Line 544: // Caches are cleared before mark-compact GC's.
Isn't this really for all mark-sweep GCs - not just the compacting ones?

http://codereview.chromium.org/457/diff/1/8#newcode557
Line 557: static Object* PutInEvalCache(bool is_global_context,
Indentation of the arguments should align them below each other.

http://codereview.chromium.org/457/diff/1/5
File src/objects.h (right):

http://codereview.chromium.org/457/diff/1/5#newcode1821
Line 1821: // EvalCache for caching eval'ed string and function.
Rename to CompilationCacheTable?

http://codereview.chromium.org/457/diff/1/4
File src/runtime.cc (right):

http://codereview.chromium.org/457/diff/1/4#newcode3331
Line 3331: bool is_global_context = context->IsGlobalContext();
I think you should move this code to the Compiler and use it for scripts too. It
would be nice to be able to just put it in compiler.cc:MakeFunction but that may
be a little to late because the Compile, CompileLazy, CompileEval functions
build script data structures and may do pre-parsing which should be avoided too.
You probably have to refactor it a bit.

[Feng Qian, 2008-09-09 03:42:50]

http://codereview.chromium.org/457/diff/1/8
File src/heap.h (right):

http://codereview.chromium.org/457/diff/1/8#newcode538
Line 538: // EvalCache caches function boilerplates for compiled scripts
On 2008/09/05 09:16:29, Kasper Lund wrote:
> I would move this code out into a separate class with static operations;
> something like CompilationCache maybe?

Let's do that in our next step. Also I'd like to see the cache performance for
whole compilation.
Another thing is that we might be able to use 1 element cache since the cached
function has source code already.
Also we may want to cache shared function info instead of the function.

http://codereview.chromium.org/457/diff/1/8#newcode544
Line 544: // Caches are cleared before mark-compact GC's.
On 2008/09/05 09:16:29, Kasper Lund wrote:
> Isn't this really for all mark-sweep GCs - not just the compacting ones?
Both, but since we use the same name Heap::MarkCompact. I probably should say
'global GC'.

http://codereview.chromium.org/457/diff/1/8#newcode557
Line 557: static Object* PutInEvalCache(bool is_global_context,
On 2008/09/05 09:16:29, Kasper Lund wrote:
> Indentation of the arguments should align them below each other.

Done.

http://codereview.chromium.org/457/diff/1/5
File src/objects.h (right):

http://codereview.chromium.org/457/diff/1/5#newcode1821
Line 1821: // EvalCache for caching eval'ed string and function.
On 2008/09/05 09:16:29, Kasper Lund wrote:
> Rename to CompilationCacheTable?

Done.

http://codereview.chromium.org/457/diff/1/4
File src/runtime.cc (right):

http://codereview.chromium.org/457/diff/1/4#newcode3331
Line 3331: bool is_global_context = context->IsGlobalContext();
On 2008/09/05 09:16:29, Kasper Lund wrote:
> I think you should move this code to the Compiler and use it for scripts too.
It
> would be nice to be able to just put it in compiler.cc:MakeFunction but that
may
> be a little to late because the Compile, CompileLazy, CompileEval functions
> build script data structures and may do pre-parsing which should be avoided
too.
> You probably have to refactor it a bit.

Done.