81ecaba321
Re-apply r20377.
...
Compute local variable liveness before translation to SSA.
Use it to remove dead values from deoptimization environments.
R=fschneider@google.com
BUG=
Review URL: https://codereview.chromium.org//14215006
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@21439 260f80e4-7a28-3924-810f-c04153c831b5
2013-04-15 11:53:50 +00:00
4d4a91ac04
Fix bug in ParallelMoveResolver::EmitSwap: implement swaps of FPU spill slots.
...
Remove representation from location. Presence of representation in location encoding was violating the invariant that unequal locations must be disjoint (where equality for locations is defined in terms of bitwise equality of their encoding). This could lead ParallelMoveResolver to treat XMM1 containing unboxed double as unequal location to XMM1 containing unboxed mint, which is obviously incorrect.
For similar reason eliminate kFloat32x4StackSlot and kUint32x4StackSlot distinction is eliminated and both are replaced with kQuadStackSlot. Register allocator now guarantees that no kQuadStackSlot occupies the same space as any other kDoubleStackSlot. This also shrinks optimized stack when only doubles are used (but might lead to a higher stack utilization when a mixture of doubles and quads is used).
Implement allocation of scratch Cpu and Xmm registers for ParallelMoveResolver. This also allows to remove push(eax)/pop(eax) pairs when resolving memory-memory cycles on ia32.
BUG=dart:9710
Review URL: https://codereview.chromium.org//13801014
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@21148 260f80e4-7a28-3924-810f-c04153c831b5
2013-04-09 12:23:46 +00:00
7e1450cb9b
Flow graph SIMD changes
...
Review URL: https://codereview.chromium.org//13471013
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@20945 260f80e4-7a28-3924-810f-c04153c831b5
2013-04-04 23:23:30 +00:00
535ed8eb25
Revert "Compute local variable liveness before translation to SSA."
...
Attaching environments to branches on strict comparisons breaks pattern matching in the optimizer and regresses performance.
This reverts commit r20377.
TBR=kmillikin@google.com
Review URL: https://codereview.chromium.org//12827027
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@20400 260f80e4-7a28-3924-810f-c04153c831b5
2013-03-22 18:32:17 +00:00
66ecd504f9
Compute local variable liveness before translation to SSA.
...
Use it to remove dead values from deoptimization environments.
R=kmillikin@google.com
BUG=
Review URL: https://codereview.chromium.org//12638040
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@20377 260f80e4-7a28-3924-810f-c04153c831b5
2013-03-22 12:33:42 +00:00
30539e6106
Register allocation tweaks:
...
- Improve heuristic detecting interference on the back-edge. Any value live_in for the loop header can introduce interfering moves at the back edge due to control flow resolution connecting split siblings. Apply adjustments based on this heuristic at the latest point when allocating a free register;
- Don't allocate a register for constants that have no constrained uses.
R=fschneider@google.com
BUG=
Review URL: https://codereview.chromium.org//12946004
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@20374 260f80e4-7a28-3924-810f-c04153c831b5
2013-03-22 11:44:38 +00:00
86da5111ad
Replace scalarlist optimizations and split external array loads into two IL instructions.
...
This CL removes optimized access for scalarlist, and only the new TypedData classes
are optimized. I changed the runtime libraries core and math to use typedData
instead of scalarlist (Uint16List is used in StringBuffer, Uint32List by Math.random).
Instead of using LoadIndexed for internal and external arrays,
split external loads into a load of the backing store and a load
of the element.
v3 <- LoadIndexed(v1, index)
becomes
v2 <- LoadUntagged(v1, ExternalTypedData::data_offset)
v3 <- LoadIndexed(v2, index);
For this I introduce two new representations in the IL:
kUntagged (for values that hold a untagged pointer) and
kNoRepresentation (for instructions accept any input
representation)
Deoptimization does not need to know about kUntagged
since these values can never occur in the environment.
Also with this change:
* fix COMPILE_ASSERT and use it in one place.
* Cleanup IL printer output of deopt ids.
Review URL: https://codereview.chromium.org//12871010
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@20198 260f80e4-7a28-3924-810f-c04153c831b5
2013-03-19 12:06:23 +00:00
9ee9cfb78d
Remove dead phis as soon as they are discovered.
...
Previously we removed dead phis late, in the register allocator. This
change removes them as soon as they are discovered to be dead and packs the
phi array to squeeze out NULLs. This speeds iteration but doesn't save
space because the phi array is zone-allocated.
The PhiIterator is used everywhere to iterate phis except a few places that
need to know the phi index (e.g., SSA construction, phi elimination).
R=vegorov@google.com
Review URL: https://codereview.chromium.org//12340108
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@19206 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-28 08:25:58 +00:00
3b374ead4b
Compile and simulate first dart function on arm generated from ast.
...
Review URL: https://codereview.chromium.org//12335102
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@19152 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-27 17:13:51 +00:00
45a6441346
Add functions for setting an environment and rebinding a use.
...
Add functions for setting or clearing an instruction's environment, which
initialize the environment uses. Add a function for changing a use's
definition.
R=vegorov@google.com
Review URL: https://codereview.chromium.org//12335063
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@19036 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-26 11:50:28 +00:00
f8d373d85b
Ignore definitions that do not have SSA index in FlowGraphAllocator::CollectRepresentations.
...
SSA building does not assign an index to definitions without uses and the rest of the allocator can cope with this treating output produced by such an instruction as temporary value.
R=srdjan@google.com
BUG=dart:8655
Review URL: https://codereview.chromium.org//12315017
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@18793 260f80e4-7a28-3924-810f-c04153c831b5
2013-02-20 22:38:22 +00:00
912194a756
Fix vm code base so that it can be built for --arch=simarm (no snapshot yet).
...
Review URL: https://codereview.chromium.org//11956004
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@17246 260f80e4-7a28-3924-810f-c04153c831b5
2013-01-18 00:34:20 +00:00
5402bf689d
Heuristically predict interference on the back edge and use it to minimize number of register reshuffling which is especially expensive when cycles of XMM registers arise.
...
When allocating free register check if hint can potentially interfere on the back edge try ignoring hint and search for a better candidate.
Additionally fix handling of constants in the liveness analysis to accommodate constants used as immediates.
R=fschneider@google.com
BUG=
Review URL: https://codereview.chromium.org//11418135
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@15291 260f80e4-7a28-3924-810f-c04153c831b5
2012-11-23 16:04:34 +00:00
78c5832342
Use structural loop info computed by register allocator to mark loop phis.
...
Currently it disagrees with loop information computed in other parts of optimization pipeline that do not require loops to be properly nested.
R=fschneider@google.com
BUG=
Review URL: https://codereview.chromium.org//11404003
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@14705 260f80e4-7a28-3924-810f-c04153c831b5
2012-11-08 19:13:57 +00:00
7c8923b829
Try allocating loop phi into a register even if phi has only unconstrained uses but there are cheap eviction candidates: values that come into the loop and have only unconstrained uses in it.
...
When spilling a value inside the loop that has only unconstrained uses in this loop move spilling point outside of the loop.
This tweaks allow to minimize amount of memory moves on loop back edges.
R=fschneider@google.com
BUG=
Review URL: https://codereview.chromium.org//11361161
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@14673 260f80e4-7a28-3924-810f-c04153c831b5
2012-11-08 02:23:41 +00:00
15114867c7
- GrowableArray::RemoveLast returns the value being removed
...
to avoid having to call Last() followed by RemoveLast().
Review URL: https://codereview.chromium.org//11348026
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@14237 260f80e4-7a28-3924-810f-c04153c831b5
2012-10-30 00:52:21 +00:00
0df78cb708
Allow splitting at lifetime holes.
...
If the allocation finger is moved past lifetime hole unwind pending interval to the start of the range before attempting to split.
R=fschneider@google.com
BUG=dart:6094
Review URL: https://codereview.chromium.org//11228025
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@13874 260f80e4-7a28-3924-810f-c04153c831b5
2012-10-22 13:36:35 +00:00
cb719805d8
Add fast 64-bit integer ADD and SUB to the optimizing compiler.
...
They operate on unboxed 64-bit integers stored in xmm registers.
I also fixed an issue with some SSE 4 specific assembler
tests where we hit an assert with an empty assembler buffer
when running with --no-use-sse41.
Review URL: https://codereview.chromium.org//11016028
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@13211 260f80e4-7a28-3924-810f-c04153c831b5
2012-10-04 08:50:28 +00:00
3ae0b96762
Set previous instruction pointers when building the graph.
...
R=kmillikin@google.com
Review URL: https://codereview.chromium.org//11043016
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@13174 260f80e4-7a28-3924-810f-c04153c831b5
2012-10-03 15:05:27 +00:00
6632f191c4
Support for unboxed 64-bit integer bitwise operations and equality on ia32.
...
This CL adds AND, OR, XOR and == operations on unboxed 64-bit integers
(aka. mints).
Unboxed mints are stored in xmm registers. Each xmm register location
has an additional bit to keep track of its value representation.
Unboxed mints are materialized on the heap on deoptmization in the same way as
unboxed doubles.
The SSE instructions used are available on all CPUs that support SSE 4.1.
Review URL: https://codereview.chromium.org//10968059
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@13112 260f80e4-7a28-3924-810f-c04153c831b5
2012-10-02 11:25:53 +00:00
7e949446ca
Disregard constants with no SSA temp index in register allocator.
...
R=fschneider@google.com
BUG=dart:5416
Review URL: https://codereview.chromium.org//10983024
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12848 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-25 16:36:56 +00:00
63fa392807
Reapply "A simpler scheme for garbage collection of ureachable phi inputs."
...
Fix two issues:
* Unreachable code elimination did not correctly identify all unreachable
blocks.
* Loop detection in the register allocator relied on the block IDs.
R=vegorov@google.com
BUG=
Review URL: https://codereview.chromium.org//10956013
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12756 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-24 07:27:36 +00:00
d2e7dba0b5
Fix bug in assigning the positions of safepoints.
...
Safepoints are associated with the end of instructions so that
splitting a live range at a safe point position makes the safepoint
at that position belong to the correct live range.
Review URL: https://codereview.chromium.org//10963029
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12708 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-21 14:07:46 +00:00
8ca2c4e6e9
Replace start_env with initial_definitions in GraphEntryInstr.
...
The initial definitions do not have any deoptimization information and can be
represented as just a list of definitions. For easy access, the constant_null
getter remains but assumes that index 0 contains the constant.
R=kmillikin@google.com
Review URL: https://codereview.chromium.org//10939031
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12555 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-19 12:37:57 +00:00
aa5281c88a
Reapply "Deoptimization support in inlined code."
...
This reapplies r12488 with a fix.
R=kmillikin@google.com ,fschneider@google.com
Review URL: https://codereview.chromium.org//10952002
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12499 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-18 13:41:20 +00:00
e174be3f27
Revert "Deoptimization support in inlined code."
...
This reverts commit a78e33f178aae74a97f4231ca98183adf97fda71.
TBR=fschneider@google.com
Review URL: https://codereview.chromium.org//10948007
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12492 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-18 12:18:04 +00:00
64a9ca2b81
Deoptimization support in inlined code.
...
Review URL: https://codereview.chromium.org//10928232
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12488 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-18 11:53:59 +00:00
5f9ff89e0a
Add support for WritableRegister policy in the register allocator.
...
Use it to eliminate push/pop on the fast path of the write barrier.
Tighten assertions in StoreIntoObject to ensure that value register is not equal to object register. Current code can't cope with that.
BUG=
Review URL: https://chromiumcodereview.appspot.com//10909169
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12193 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-11 12:50:52 +00:00
8201f0db6b
Introduce nested deoptimization environments.
...
Add an optional outer deoptimization environment to support inlining.
Introduce a pair of environment iterators.
Original codereview: https://chromiumcodereview.appspot.com/10928048/
R=fschneider@google.com ,zerny@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10909168
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12190 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-11 12:30:21 +00:00
7cc84c2892
Implement new optional parameters syntax in the vm (issue 4290).
...
Provide --reject_named_argument_as_positional flag to ease transition to new
syntax (default is false).
Add tests.
Review URL: https://chromiumcodereview.appspot.com//10910119
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@12004 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-07 01:13:27 +00:00
a192ac3d9e
Allow smi comparisons to have constant operands.
...
Remove unnecessary temps reserved for them.
Load constants directly in phi moves.
Generate xor(reg, reg) for loading of 0 instead of mov.
R=fschneider@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10918097
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11947 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-06 13:27:30 +00:00
da14bf70a7
Add attributions so printf like functions can have their arguments checked.
...
This change also corrects some misuses of format strings and format
arguments that discovered by the compiler checks.
Review URL: https://chromiumcodereview.appspot.com//10869063
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11912 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-06 00:58:25 +00:00
0750c472dc
Reapply "Remove classes Computation and BindInstr."
...
Reapply svn revision 11868 with a bugfix.
Original code review at http://codereview.chromium.org/10908091/
R=fschneider@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10912094
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11890 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-05 17:04:49 +00:00
7e36720fa8
Revert "Remove classes Computation and BindInstr."
...
This reverts svn revision 11868.
TBR=fschneider@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10917085
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11869 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-05 11:20:14 +00:00
bd84c7ee22
Remove classes Computation and BindInstr.
...
The former Computation classes are folded into the Instruction class
hierarchy.
R=fschneider@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10908091
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11868 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-05 11:15:34 +00:00
1911b2bc81
Make register allocator to disregard constant computation with no uses.
...
Inline constant values into environments, instead of creating a use of the constant definition.
Change BinarySmiOpComp code template to allow inline constant uses for the right hand side.
R=fschneider@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10916082
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11798 260f80e4-7a28-3924-810f-c04153c831b5
2012-09-04 09:58:57 +00:00
a573d2dec8
Unbox phis that were proven to be of type Double.
...
Eliminate Boxing/Unboxing pairs.
Allow boxing, unboxing and double binary operations to participate in CSE.
Allow double comparisons to operate on unboxed inputs.
Support XMM registers and double spill slots in deoptimization.
Save XMM registers when calling to runtime from WriteBarrier stub.
R=srdjan@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10919008
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11652 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-30 20:39:48 +00:00
978ec58a5e
Rematerialize constants instead of spilling them.
...
This also fixes the crash that occurs when we try to spill constant null at the graph entry.
R=srdjan@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10918006
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11634 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-30 17:49:23 +00:00
8122966c80
Put live register bits in stackmaps.
...
Add a count of live registers to each stackmap and add bits describing the
registers. This allows untagged values in general purpose registers at
safepoints.
R=vegorov@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10882055
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11595 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-30 10:20:26 +00:00
ea925449f6
Eliminate class UseVal.
...
This was the only subclass of Value. All functionality is moved to the base
class, and all functions are made non-virtual.
R=fschneider@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10891025
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11590 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-30 09:08:27 +00:00
58d842d6d7
Make constants computations instead of values.
...
This change requires a pool of global constants that don't have an easy
place to go in the instruction stream. Currently, that is only the constant
null used to mark eliminated phis in deoptimization environments.
R=fschneider@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10894034
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11518 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-29 12:56:57 +00:00
264c4f6d1a
Add support for XMM registers in SSA code generation pipeline.
...
Split BinaryDoubleOp into several instructions that manipulate unboxed doubles.
R=fschneider@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10875030
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11313 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-24 14:45:42 +00:00
323e06222e
Reland "Add a simple dominator based redundancy elimination" and fix a register allocation bug.
...
Original CL: http://codereview.chromium.org/10872035/
It enables elimination of redundant expressions across
basic blocks.
Currently used for smi checks and class checks, but will
be extended to other expressions in a separate CL.
Additionally, this fixes a bug in the register allocator where
a blocked register was illegally assigned to an unallocated live range.
It also fixes a Mac compiler issue with the constructor of
DirectChainedHashMap.
Review URL: https://chromiumcodereview.appspot.com//10871060
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11302 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-24 11:10:36 +00:00
50b114c606
Compute the def-use list on-demand by walking the dominator tree.
...
This replaces the previous implementation that maintained the def-use lists.
R=vegorov@google.com ,kmillikin@google.com
Review URL: https://chromiumcodereview.appspot.com//10879036
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@11241 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-23 16:23:05 +00:00
6ed5884c44
RemoveFromGraph on definitions asserts empty use lists.
...
R=fschneider@google.com ,kmillikin@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10855101
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@10822 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-16 14:08:45 +00:00
5cc78d2a33
Refactored FlowGraphBuilder into a separate FlowGraph representation.
...
R=kmillikin@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10857016
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@10807 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-16 12:06:17 +00:00
d571e18ebd
Make all variants of RelationalOp work with SSA.
...
This is the last instruction that causes a bailout in the register
allocator. This bailout is now changed into an ASSERT.
Review URL: https://chromiumcodereview.appspot.com//10830333
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@10802 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-16 08:46:45 +00:00
181be0165b
Rename is_call to always_calls and contains_call to can_call.
...
R=kmillikin@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10824308
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@10735 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-15 14:26:39 +00:00
fe291bf0fa
Implement basic support for deferred slow path code with calls that save and restore live registers.
...
Use it to avoid spilling over stack-checks.
R=kmillikin@google.com
BUG=
Review URL: https://chromiumcodereview.appspot.com//10823308
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@10639 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-14 12:47:42 +00:00
172ee045f3
Build and use stack maps in the SSA compiler.
...
Add a stack bitmap to the location summaries for calls that are GC
safepoints. The bitmap covers the spill slots. The register
allocator collects these bitmaps into a list and then marks live
pointer values during register allocation. When emitting code for a
call, a heap-allocated stackmap is built.
BUG=
Review URL: https://chromiumcodereview.appspot.com//10831261
git-svn-id: https://dart.googlecode.com/svn/branches/bleeding_edge/dart@10618 260f80e4-7a28-3924-810f-c04153c831b5
2012-08-14 06:59:24 +00:00
vegorov@google.com